Architecture remodeling activates the HerA-DUF anti-phage defense system

Summary

Leveraging AlphaFold models and integrated experiments, we characterized the HerA-DUF4297 (DUF) anti-phage defense system, focusing on DUF’s undefined biochemical functions. Guided by structure-based genomic analyses, we found DUF homologs to be universally distributed across diverse bacterial immune systems. Notably, one such homolog, Cap4, is a nuclease. Inspired by this evolutionary clue, we tested DUF’s nuclease activity and observed that DUF cleaves DNA substrates only when bound to its partner protein HerA. To dissect the mechanism of DUF activation, we determined the structures of DUF and HerA-DUF. Although DUF forms large oligomeric assemblies both alone and with HerA, oligomerization alone was insufficient to elicit nuclease activity. Instead, HerA binding induces a profound architecture remodeling that propagates throughout the complex. This remodeling reconfigures DUF into an active nuclease capable of robust DNA cleavage. Together, we highlight an architecture remodeling-driven mechanism that may inform the activation of other immune systems.

In brief

HerA-DUF, an anti-phage defense system, consists of HerA (ATPase) and DUF (nuclease). The complex formation of HerA-DUF reorganizes the DUF dodecamer, triggering nuclease domain dimerization and activation, leading to effective anti-phage defense.

Graphical Abstract

INTRODUCTION

The recent advent of artificial intelligence (AI)-driven structure prediction tools, particularly AlphaFold, has transformed structural biology and protein biochemistry^1,2. These methods have generated an expansive repository of highly accurate three-dimensional (3D) protein models^1,3. Given the pivotal role of 3D structures in dictating protein function, these predictive models provide unprecedented opportunities to elucidate the roles of previously enigmatic proteins.

The arms race between phages and bacteria has prompted the development of diverse anti-phage defense mechanisms in bacteria^4–6. Bioinformatic analyses have uncovered numerous bacterial immune systems, many of which harbor components with elusive functions^7,8. One such system, HerA-DUF4297, comprises HerA, a well-characterized ATPase, and DUF4297 (DUF), a protein with undefined biochemical function⁸ (Figure 1A). HerA, an ATPase with helicase activity, widely exists in bacteria and archaea. For example, HerA, together with a nuclease called NurA, has been shown to form a stable complex for processing DNA double-strand breaks ⁹. DUF is predicted to belong to the PD-DExK-nuclease superfamily through sequence analysis⁸. Members of this family are widespread in bacteria immune defense systems, highlighting their functional importance. Prior phage infection assays have shown that both HerA and DUF are required for effective anti-phage defense, yet the underlying mechanisms remain unknown⁸.

Figure 1. Integrated functional characterization of HerA-DUF.

(A) Schematic diagrams of DUF and HerA with domains color coded.

(B) Workflow of integrated mechanistic and functional analysis of DUF. AlphaFold predicted structure of DUF was used to identify its structural homologs using Foldseek. Based on the functions of Cap4, a DUF structural homolog, we designed biochemical and structural experiments to elucidate functions of DUF.

(C) HerA-DUF, but not DUF alone or HerA alone, displays robust nuclease activity. The gel is a representative of assays performed in triplicate.

(D) Phylogenetic tree of DUF with N = 1009 DUF nuclease domain-containing sequences with some major conserved clusters highlighted based on shared biological function. The tree is configured in a polar tree layout, rooted at the midpoint with the root hidden, nodes set in increasing order, and branches transformed in cladogram style for ease of visualization. Cluster 2 (AVAST3b family) has poorly conserved sequences and spans roughly 70% of the tree.

(E) 2D genomic context cluster map, generated by GCsnap, with DUF nuclease domain-containing proteins clustered by biological function. Clusters associated with similar biological functions or protein classes are boxed together to aid visualization.

(F) Cartoon representation of gene families within each identified cluster occurring with ≥ 35% frequency. Each DUF nuclease domain-containing gene is colored grey. Clusters are grouped into associated functional families by dashed lines.

(G) HerA-Duf displays robust nuclease activities towards 50 bp dsDNA in the presence of Mn²⁺ but not other cations tested. In contrast, HerA-Duf can effectively process DNA substrates of 1,000 bp in the presence of Mg²⁺ or Mn²⁺ with higher cleavage efficiency for Mn²⁺.

(H) HerA-DUF can eliminate plasmids in supercoiled, circular, and linear forms. In contrast, apo DUF can only nick supercoiled into circular plasmid but cannot cleave linear plasmid, indicating a low nuclease activity. The gels shown are representative of assays performed in triplicate.

See also Figure S1

Here, we integrate data mining, structural elucidation, and biochemical assays to delineate the function of the HerA-DUF anti-phage defense system. Leveraging the AlphaFold-predicted structure of DUF, which comprises an N-terminal domain (NTD) and a C-terminal domain (CTD), we employed Foldseek to identify structural homologs of DUF¹⁰ (Figure 1B). This search revealed a striking similarity between the DUF NTD and the nuclease domain of Cap4, despite modest primary sequence identity¹¹. Guided by this structural parallel, our biochemical experiments confirmed that DUF itself is inactive as a nuclease, whereas the HerA-DUF complex displays robust DNA-cleavage activity.

Further cryo-EM structural analyses depict distinct architecture remodeling of DUF alone and in complex with HerA. While DUF alone forms compact oligomers with monomeric nuclease domains, its direct association with HerA induces the rearrangement of DUF CTD, leading to proximity-induced formation of distinctive “clamp-shaped” dimers of DUF nuclease domains ideally suited for substrate DNA engagement. This architecture remodeling, rather than oligomerization per se, underpins DUF’s transition into an active nuclease.

Together, these findings highlight an integrated approach to study protein function, uncover dramatic architecture remodeling of DUF, and challenge the prior assumption that protein oligomerization by itself is sufficient to drive immune signaling.

RESULTS

Structure-guided clustering of DUF nuclease domain-containing proteins

To characterize the structure and function of DUF, we first employed the AlphaFold-predicted structure of the DUF nuclease domain as a template to identify structural homologs with Foldseek to guide our biochemical analyses^10,12 (Figure 1B and 1C). Based on the Foldseek search against the AlphaFold database (AFDB), we selected 48 protein matches from AFDB, including a Cap4 endonuclease with low sequence identity (10%), for genomic characterization of DUF homologs^11,13. We conducted PSI-BLAST analyses to identify sequence homologs of the 48 targets¹⁴, and assembled a dataset of 1,009 DUF nuclease-containing proteins for further analyses. To examine the structural and functional diversity of DUF homologs in bacterial immunity, we used 1,000 representative DUF homologs to generate a phylogenetic tree, revealing several major sequence-based groups (Figures 1D and 1E). Among those major groups, we found known bacterial immune systems, such as CBASS and DEAD/DEAH box helicases-associated systems⁴ (Figures 1D and 1E). In contrast, AVAST systems were distributed among multiple groups, indicating notable sequence diversity ⁸ (Figures 1D and S1A). The HerA-DUF system emerged as phylogenetically distinct from these other clusters, suggesting a more distant evolutionary origin (Figures 1D and 1E).

Further clustering of all the DUF homologs with CLANS until equilibrium generated 18 clusters¹⁵ (Figure 1E). The majority of these clusters can be further consolidated into five major families encompassing known bacterial immune systems, including CBASS, DEAD-Box helicase-associated proteins, NACHT, and AVAST type 3b family of proteins (Figure 1F). Cluster 3 specifically contained the HerA-DUF system (Figure 1F). We then explored the defense islands, or the surrounding genetic loci of these clustered sequence groups with GCsnap, and identified some conserved gene families within the clustered groups¹⁶ (Figure 1F). Cluster 3 stood out by lacking any additional conserved gene families beyond HerA and DUF (Figures 1F and S1A). In contrast, groups associated with CBASS and DEAD Box helicase groups were accompanied by a variety of associated conserved families located within the genetic loci, underscoring diverse functional roles of DUF nuclease domain homologs in different anti-phage systems (Figure 1F).

Together, these analyses showed that DUF nuclease domain-containing proteins, with diverse sequences and structures, are widespread in diverse bacterial immune systems, underscoring their functional importance in anti-phage immune defense.

DUF-HerA promotes nuclease activity of DUF

Given the structural similarity between DUF and Cap4, a metal ion-dependent endonuclease¹¹, we next assessed the potential nuclease activity of DUF and its metal preference using various DNA substrates (Figures 1C, 1G and S1B–D). We found that HerA-DUF, but not DUF alone, can effectively process short double-stranded DNA (dsDNA) in the presence of manganese, but not with other divalent cations tested (Figures 1G and S1B–D). In contrast, HerA-DUF can process long dsDNA in the presence of either magnesium or manganese. Since HerA-DUF, but not DUF alone, can process long dsDNA in the presence of magnesium (Figure 1C), we further tested whether this holds true for plasmid DNA. We found that HerA-DUF can digest plasmid DNA, whereas DUF alone only generated a nicked product, underscoring the necessity of HerA-DUF complex formation for robust nuclease activity (Figure 1H). Consistently, HerA-DUF can effectively process linearized plasmid DNA, whereas DUF alone failed to do so (Figure 1H). Together, these data establish that forming a complex with HerA is critical for DUF activation.

Structure of apo DUF

Since oligomerization is essential for activating many nucleases, including Cap4¹¹, we investigated whether DUF assembles as an oligomer. Unexpectedly, DUF alone displayed a relatively broad peak on gel filtration, indicative of heterogeneous oligomeric states (Figure S2A). Mass photometry further confirmed that DUF populates a range of oligomers, including dimers, hexamers, octamers, decamers, dodecamers, and tetradecamers (Figure 2A).

Figure 2. Apo DUF forms oligomers.

(A) DUF exists as variable oligomers revealed by mass photometry.

(B) Cryo-EM map of apo DUF dodecamer.

(C) Surface representation of apo DUF dodecamer from the top and side views.

(D) Ribbon diagram of apo DUF dodecamer from the side view.

(E) Ribbon diagram of DUF top (green) and bottom (cyan) layers.

(F) DUF oligomerization is mediated by the CTD domain.

See also Figures S2 and S3, Tables S1 and S2.

To visualize these assemblies, we performed cryo-EM analyses and determined the structures of apo DUF, revealing that DUF alone indeed forms various oligomers, consistent with our gel filtration and mass photometry findings (Figures S2A–G). We identified tetramers, hexamers, and dodecamers in our structural analyses, with dodecamers being the dominant species (Figures S2B and S2F). The dodecameric DUF was reconstructed resolved to a nominal resolution of 3.4 Å while others were determined to resolutions of 3.7–3.8 Å (Figures S2F and S3A). In particular, the resolution of DUF CTD in the dodecamer was further optimized to 3.2 Å through local refinement, enabling us to precisely build an atomic model (Figures S2F–H). As predicted, each DUF protomer adopts a bipartite fold with an N-terminal nuclease domain and one C-terminal domain. Given the dominance of the dodecameric state, we focused on this assembly for elaborating apo DUF assembly below.

The overall structure of apo DUF is composed of two layers with six protomers per layer (Figures 2B–D). Notably, the six protomers in each layer form an open-ring structure through extensive interactions with neighboring subunits (Figure 2E). The two layers are linked together through interactions mediated by one protomer from the top layer and one protomer from the bottom layer in a shoulder-to-shoulder manner (Figure S3B). The CTDs packed tightly against each other to form the core with the nuclease domains flanking around. At the center of the DUF core, 12 copies of CTDs form extensive interactions to assemble into a cylinder-shaped architecture (Figure 2F).

Each protomer of the DUF CTD adopts an α-β-α Rossmann fold with a six-stranded β-sheet surrounded by six α helices (Figure S3C). As Rossmann fold exists widely in many different enzymes, we found more than 100 homologs of DUF CTD in PDB database through DALI search. Among these homologs, the large subunit of prokaryotic SPARSA SIR2 domain and the N-terminal domain of N-acetylglutamate synthase^17,18 are the top two candidates with root mean square deviations (RMSDs) of 3.3 and 3.5 Å, respectively (Figure S3D). Notably, the CTD of DUF, the SIR2 domain of prokaryotic SPARSA, and the N-terminal domain of N-acetylglutamate synthase can all mediate protein oligomerization^17,18. As such, we denoted this domain as a Rossmann Fold for Oligomerization (RFO) domain.

Assembly of HerA and HerA-DUF

To understand how HerA interacts with DUF, we first determined the cryo-EM structure of HerA at 3.26 Å, revealing that HerA forms an asymmetric hexamer (Figure 3A, S4A–D). HerA is composed of three domains: a helical hairpin (HAS) domain at the N-terminus, a RecA-like ATPase domain, and an inserted helical bundle domain, followed by a C-terminal brace (Figure 3B). Notably, the HerA we studied contains a long C-terminal brace, contrasting with the HerA molecule in the HerA-Sir2 complex^19,20 (Figure S4E).

Figure 3. Assembly of HerA and HerA-DUF.

(A) Structure of one HerA protomer with domains color coded.

(B) Ribbon diagram of apo HerA hexamer with side and top views.

(C) Structure of HerA-DUF complex in ribbon diagram with HerA and DUF color coded.

(D) Top view and side view of HerA in complex with DUF RFO domains. A central channel runs through the entire HerA-DUF complex with a large opening at the top of DUF.

(E) Interfaces mediating interactions between HerA HAS domain (magenta) and DUF RFO domain (cyan). All six protomers of HerA interact with DUF through a major interface. However, only one protomer of HerA out of six interacts with a secondary RFO of DUF through a minor interface.

(F) Diagram illustrating interactions between HerA and DUF. The six protomers for HerA are denoted “a-f ” and the six protomers of bottom layer DUF are denoted “A-F.”

(G)HerA E92K and E25K mutants substantially reduced the nuclease activity of HerA-DUF.

(H) Upon forming a complex with DUF, HerA undergoes dramatic rearrangement in comparison to the apo HerA.

See also Figure S4 and Table S1.

We next solved the cryo-EM structure of the HerA-DUF complex at a resolution of 2.76 Å (Figures S2D–G). Within the complex, 12 copies of DUF assemble into two layers on top of the HerA hexamer (Figure 3C). The bottom layer of DUF, but not the top layer, forms extensive interactions with the HerA hexamer, with a total buried interface of ~3,100 Ų (Figures 3D–E). Specifically, HerA uses its N-terminal HAS domain to interact with the RFO domains of the bottom DUF layer, similar to other HerA proteins that engage partner proteins via their HAS domains (Figures 3D–3F). However, compared to the HAS domain in other HerA proteins, our structure reveals unique inserted motifs and interaction hotspots (Figures S4F and S4G).

In the DUF-HerA complex, two interfaces—major and minor—mediate interactions between the HerA HAS domain and the DUF RFO domain (Figures 3E and 3F). The major interface facilitates paired interactions between all the protomers at the bottom layer of DUF and the six protomers of HerA (Figure 3F), while the minor interface occurs between one HerA protomer out of six and one DUF protomer out of six (Figure 3F). The major interface, with a buried surface area of ~480 Ų, is formed by loops from both the HAS domain and the RFO domain, whereas the minor interface of ~240 Ų involves a short α-helix of the HAS domain and loops of the RFO domain (Figures 3F, S4H, and S4I). Both interfaces are dominated by hydrophilic and charged residues (Figures S4H and S4I).

Interactions between HerA and DUF regulates their activities

Interactions between HerA and DUF promote DUF nuclease activity. Biochemical analysis showed that the HerA-DUF complex, but not DUF alone, can effectively cleave DNA substrates. To test the functional importance of HerA-DUF interactions in catalysis, we mutated HerA residues critical for DUF binding, including E92K, E25K, and H44D (Figures S4H and S4I). These mutants affected the complex formation, as evidenced by altered gel filtration profiles (Figures S4J–L). Further nuclease activity assay revealed that the mutant HerA-DUF complexes have much lower nuclease activities than wild type, highlighting the functional importance of HerA-DUF interactions in promoting the DUF’s nuclease activity (Figure 3G).

Interactions between DUF and HerA reshaped the tetrameric structure of HerA hexamer (Figures 3H, S4M, and S4N). Structural comparison between apo HerA and HerA in the complex revealed four distinguished features. First, the HerA in the complex has a C3 symmetry while the apo HerA is asymmetric (Figure 3H). Second, the helical bundle domains of HerA in the complex packed tightly together, forming a narrow pore with a diameter of ~15 Å in the middle (Figure 3H). In contrast, the helical bundle domains of apo HerA are positioned farther apart, resulting in a wider pore of ~30 Å in diameter (Figure 3H). Third, the HAS domain in the complex forms a wider pore with a diameter of 40 Å, compared to a narrower pore with a diameter of 20 Å in the apo HerA (Figure S4M). Fourth, HerA protomers in the HerA-DUF complex form more extensive interactions with each other with a total buried surface area of ~19,940 Ų, compared with a total buried area of ~18,120 Å in the apo HerA hexamer (Figure S4N).

The assembly of HerA-DUF modulates the ATPase activity of HerA (Figure S4O). HerA alone displayed modest ATP hydrolysis activity. When forming a complex with DUF, the ATPase activity of HerA was substantially promoted (Figure S4O). Perhaps, the architecture remodeling of HerA hexamer, when forming a complex with DUF, contributed to the enhanced ATP hydrolysis.

As HerA often functions as a helicase ⁹, we tested the helicase activity of HerA using a forked DNA substrate. While the positive control, DDX43, displayed robust helicase activity ²¹ (Figure S4P), neither HerA alone nor the HerA-DUF complex can unwind DNA (Figure S4P), indicating that HerA alone or in complex with DUF has no helicase activity.

Additionally, we found that ATP inhibits the nuclease activity of the HerA-DUF system, with 0.5 mM ATP sufficient to suppress its nuclease activity (Figure S4Q). This inhibition appears uncoupled from HerA’s ATP hydrolysis, , as both wild-type and ATPase-deficient HerA–DUF complexes showed similar nuclease activity and were equally inhibited by 0.5 mM ATP(Figures S4Q and S4R). These observations raise the possibility that ATP may directly inhibit DUF.

RFO domain drives the formation of inactive and active DUF oligomers

Within apo DUF, 12 protomers of DUF form a bilayer structure with six protomers in each layer, like a cylinder (Figures 4A–C). Four different interfaces are identified in the dodecamer of apo DUF, denoted as Interface I to Interface IV (Figure 4C). Interface I mediates interactions within each layer with a buried area of ~ 500 Ų (Figures 4C and S5A). Interface II and Interface III contribute to the assembly of protomers between layers with buried areas of ~1,260 Ų and ~570 Ų, respectively (Figures 4C, S5B, and S5C). Detailed analysis showed that all three interfaces are dominated by hydrophilic and charged residues (Figures S5A–S5C). In contrast, interface IV is mainly contributed by hydrophilic residues at the linker region with a buried area of ~470 Ų, which seals the two open-rings on the top and at the bottom layers (Figures S3E and S5D).

Figure 4. RFO drives the formation of divergent DUF oligomers.

(A) Apo DUF assembles into a cylindric architecture with two layers, mediated by the RFO domain interactions.

(B) Protomers in the top and bottom layers of apo DUF form an open-ring structure with an inner diameter of 85 Å.

(C) Four types of interfaces mediate the assembly of RFO domains in apo DUF. Type I exists between adjacent subunits within each layer, while type II and type III interfaces mediate interactions between the top and bottom layers. Type IV is uniquely positioned to close the top and bottom ring.

(D) Active DUF assembles into a cone-like structure with a wide bottom and a narrow top through RFO-RFO interactions.

(E) The top layer of active DUF RFO domains form a closed hexameric ring with a central pore, while the bottom layer forms a dispersed hexameric ring.

(F) Two types of RFO interfaces, comparable to type I and type II in apo DUF, dictate the assembly of active DUF.

See also Figure S5.

Within the HerA-DUF complex, DUF exists as an active state and assembles as a dodecamer through RFO-RFO interactions (Figures 4D–4F). Similar to the apo structure, the 12 protomers of DUF in the active state form a bilayer structure with six protomers in each layer (Figures 4D and 4E). However, unlike the open-ring structure of the apo state, the six protomers in the top layer form a closed ring through extensive interactions with neighboring subunits, while the six protomers in the bottom layer do not interact with each other (Figures 4E and 4F). Consequently, the RFO domains in active DUF resemble a cone, starkly contrasting the cylindrical structure observed in the inactive state (Figures 4A and 4D). These architectural differences originate from interactions between DUF and HerA, leading to dramatic changes in the interaction interfaces of DUF oligomers (Figures 3C, 4C, and 4F). In active DUF oligomers, two distinct interfaces, corresponding to Interfaces I and II in the apo state, are observed (Figures 4F, S5E, and S5F). Interface I, with a buried area of ~470 Ų, mediates interactions between neighboring subunits in the top layer, while Interface II, with a buried area of ~830 Ų, facilitates interactions between the top and bottom layers (Figures 4F, S5E, and S5F). Furthermore, the total buried areas in the active DUF are substantially smaller than those in the apo state. These differences dictate the topological variations of RFO in active versus apo DUF, which propagate to the nuclease domain, leading to the topological divergence of the nuclease domains between active and apo DUF.

Assembly of the RFO domain is critical for the activation of DUF. Substitution of interfacial residues can effectively disrupt the oligomerization of DUF as determined by gel filtration analysis (Figure S5G). These interfacial mutants also affected the assembly of the HerA-DUF complex (Figure S5H). These interfacial mutants substantially reduced the nuclease activity of the HerA-DUF complex (Figure S5I).

Together, our structural analysis underscores the critical role of the RFO domain in driving the divergence of DUF oligomers across different states.

Nuclease domain dimerization activates DUF

The architecture of active DUF exhibits dramatic differences from that of apo DUF, revealing the process of DUF activation (Figures 5A and 5B). In the apo DUF dodecamer, protomers in the top and bottom layers are orientated clockwise and counterclockwise to each other, leading to the tight packing of the nuclease domains of DUF (Figure 5A). In contrast, the conformations of the DUF protomers in the active state undergo a global rearrangement, revealing a mechanism for the activation of DUF. First, the bottom layer of active DUF displayed a dispersed conformation (Figure 5B). Second, the neighboring protomers of DUF on the top layer form a swapped dimer (Figures 5B and S6A–D). The initial map of the top layer is very poor without much density (Figure S2F). We performed symmetry expansion to increase the particles of active DUF for local refinement and substantially improved the density of DUF top layer (Figures S2F and S6B–D). Though the nuclease domain resolution is not sufficient for building an atomic model, we are confident to fit the AlphaFold predicted DUF into to the density (Figures S6C–D). In total, there are three pairs of DUF dimers on the top layer (Figure S6B). These dimers can form between any adjacent DUF protomers on the top layer, resulting in two possible configurations of three pairs of dimers (Figure S6A). Together, these data revealed a mechanism of global domain rearrangement during the switching of DUF from apo state to active state.

Figure 5. Dimerization of nuclease domain activates DUF.

(A) DUF nuclease domains in each layer of apo DUF rotate anti-parallel to each other and pack asymmetrically without contacting each other.

(B) Nuclease domains in each layer of active DUF display distinct rearrangements. In the top layer, two adjacent nuclease domains (orange and green) form a dimer. In contrast, nuclease domains in the bottom layer (blue) adopts a dispersed ring-like structure.

(C) All the DUF protomers in apo state share the same conformation, while there are three conformations observed in active DUF protomers.

(D) Overlaid protomers of DUF in apo and active state, revealing their domain rearrangement across states.

(E) Structure of DUF dimer in complex with DNA (purple) and a blowup of the DUF active site with magnesium, predicted by AlphaFold 3.

(F) Substitutions of key residues mediating DUF nuclease domain dimerization substantially reduced nuclease activity of HerA-DUF. The gels shown are representative of assays performed in triplicate.

(G) Substitution of catalytic residues abolished nuclease activity of HerA-DUF. D41A* represents a rescuing experiment by adding additional magnesium, indicating that D41 is responsible for coordinating magnesium. The gels shown are representative of assays performed in triplicate.

See also Figure S6.

The architecture remodeling of DUF is driven by rigid body rotation of the nuclease domain, accompanied by pronounced structural changes in the linker region between the nuclease domain and the RFO domain (Figures 5C–5E and S6E–G). The protomer of apo DUF adopts a folded conformation (denoted as DUF^Apo), whereas protomers of active DUF have three conformations (Figure 5C). As such, we denoted protomers at the bottom layer as DUF^AB, protomers on the top layer as DUF^AT1 and DUF^AT2 due to their conformational differences in the linker region. The conformation of DUF^AB resembles the conformation of DUF^Apo with limited tilting in the nuclease domain (Figure S6E). However, when aligned with DUF^Apo using the RFO domain, the nuclease domains of DUF^AT1 and DUF^AT2 rotate about 75° and 140°, respectively, along an axis at the junction between the nuclease domain and the RFO domain for two neighboring protomers on the top layer (Figure 5D). The structural change is approximately a rigid body rotation as the RFO domain and the nuclease domain align well in gross conformations between the inactive and the active states (Figures S6F and S6G). Strikingly, the linker region between the nuclease domain and the RFO domain in protomers of the top layer of active DUF undergoes dramatic conformational changes, compared to that in the apo DUF (Figure S6H). The conformational changes in the linker region include secondary structure changes from loops into alpha helices and changes from folded conformation to a stretched conformation (Figure S6H). For the protomers at the bottom layer of active DUF, the linker region adopted a similar conformation with that in the inactive DUF but titled a certain angle, leading to an extended conformation of the nuclease domain (Figures 5D and S6H). Together, these data revealed the structural basis of DUF architectural rearrangement from an inactive state to an active state.

Dimerization of the DUF nuclease domain is likely crucial for binding and cleaving DNA substrates. This dimerization is mediated not only by the RFO domains but also by linker-linker and nuclease-nuclease interactions, with a similar topology to the nuclease domain dimer of AVAST3 ²² (Figures 5E, S6I, and S6J). Hydrophilic residues dominate the interactions in the linker region, while hydrophobic residues facilitate the interactions between neighboring nuclease domains (Figures S6K and S6L). Together, these interactions create a buried interface of ~1,500 Ų, stabilizing the dimeric DUF nuclease domain for effective DNA binding. The AlphaFold3-predicted structure of the DUF-DNA complex reveals that double-stranded DNA is coordinated in the center of the clamp-shaped DUF dimer² (Figure 5E). Key residues, including hydrophilic and positively charged ones, are critical for this coordination. Consequently, the dsDNA is well-coordinated by the clamp-shaped nuclease domain, with two strands poised for processing by the two protomers, explaining the robust activity of the DUF-HerA complex (Figure 5E). Consistent with our structural analysis, substitutions of key residues critical for mediating nuclease domain dimerization substantially reduced the nuclease activities of DUF, highlighting the importance of nuclease domain dimerization in DNA cleavage (Figures 5F, S6K, and S6L).

The active site of the DUF nuclease domain is composed of a catalytic triad, resembling that of the Cap4 nuclease domain¹¹ (Figure 5E). Based on the AlphaFold3 prediction², D41 is responsible for coordinating magnesium while catalysis also requires E59 and K61 (Figure 5E). Substitutions of these residues abolished the nuclease activity of DUF, underscoring the functional importance of these residues in catalysis (Figure 5G).

Together, these results uncover the mechanism of DUF activation driven by dimerization of the DUF nuclease domain.

The central pore of DUF oligomers coordinates DNA substrates

To understand how DNA substrates were coordinated by HerA-DUF, we incubated 50 bp DNA substrate with the HerA-DUF complex and determined the cryo-EM structure of the DUF-HerA-DNA ternary complex to a resolution of 3.0 Å (Figures 6A, S7A, and S7B). An additional density for dsDNA was clearly visible in the central pore of DUF oligomers though it did not permit an atomic-level model (Figures S7C-D). Therefore, we fitted a short dsDNA into the density. Consistently, the rearrangement of DUF from inactive state to active state aligns positively charged residues in the central pore of active DUF, facilitating DNA binding. (Figures 6B and 6C). Further EMSA assay showed that HerA alone cannot bind to DNA (Figure 6D). In contrast, both HerA-DUF and DUF can bind to DNA with HerA-DUF developing a higher binding affinity, supporting that the DUF rearrangement within the HerA-DUF complex indeed facilitates DNA binding (Figure 6D). When the positively charged residues in the central pore of DUF were substituted by glutamate, the mutants failed to bind DNA, highlighting our structural observation that the central pore of DUF is responsible for DNA coordination (Figure 6E). Consistently, these DNA binding defective mutants displayed reduced nuclease activities (Figure 6F), highlighting their functional importance in coordinating DNA substrates to promote catalysis. Our phage plaque assays showed that the DNA binding defective mutant HerA-DUF complexes have reduced anti-phage defense activities in bacteria, highlighting the functional importance of DNA coordination by DUF in anti-phage defense (Figure 6G).

Figure 6. DNA coordination by HerA-DUF.

(A) Cryo-EM density of DUF-DNA complex, revealing that DNA binds to the central pore of DUF.

(B) Electrostatic surface of active DUF RFO domains, revealing an aligned pattern of positive-charged residues in the central pore.

(C) Electrostatic surface of apo DUF RFO domains, revealing a dispersed pattern of positive-charged residues in the central pore.

(D) The HerA-DUF complex likely binds dsDNA more efficiently than apo DUF, as revealed by gel shift assay. In contrast, HerA alone appears not to bind dsDNA. The gel shown is representative of assays performed in triplicate.

(E) Substitutions of positively charged residues in the central pore of DUF abolished DNA binding, supporting the critical role of DUF central pore in coordinating DNA.

(F) Substitutions of positively charged residues in the central pore of DUF reduced nuclease activities of DUF.

(G) Antiphage plaque assays show that HerA or DUF mutants display reduced activities. Among these mutants, DUF mutants are highlighted in blue while HerA mutants are highlighted in pink. DUF D412R, F13/17A, and D206K affect the oligomerization of DUF. K265/314/434E is a central pore triple mutant, abolishing DNA binding by the DUF central pore. K56/125E is a mutant that affects DNA binding to the active site of DUF. D41A is a catalytic dead mutant. HerA E438/439A is an ATP hydrolysis defective mutant, whereas E92K affects the complex formation of DUF-HerA. Image is representative of assays performed in triplicate.

See also Figure S7 and Table S2.

Together, the central pore of active DUF developed better binding affinity towards DNA substrates, promoting nuclease and anti-phage defense activities of the Her-DUF complex.

Architecture remodeling activates DUF

Our structural and biochemical analyses uncovered an intriguing mechanism for the activation of DUF that is mediated by global architectural remodeling upon complex formation with HerA (Figures 7A and 7B). In the apo state, DUF predominantly assembles into double-layered oligomers through RFO-mediated interactions, with each nuclease domain existing as a monomer (Figure 7A). Interactions with HerA induce substantial rearrangements of DUF, in particular the RFO domains of DUF transition from a cylindrical to a cone-shaped architecture (Figure 7B). This rearrangement of the RFO in active DUF leads to the proximity-induced dimerization of nuclease domains at the top layer (Figure 7A). Dimerized nuclease can form an active site for processing dsDNA. Substitutions of key residues for nuclease dimerization reduced the nuclease and anti-phage activities of the HerA-DUF complex (Figure 6G). Together, our study revealed that the activation of DUF required a few steps, including DUF oligomerization, HerA-DUF complex formation, architecture remodeling of DUF oligomers, and the dimerization of the DUF nuclease domain (Figure 7A).

Figure 7. Architecture remodeling activates HerA-DUF.

(A) RFO domain drives the formation of distinct DUF oligomers across different states. Upon forming a complex with HerA, DUF undergoes dramatic architecture remodeling, leading to the activation of DUF.

(B) Upon interacting with HerA, the rearrangement of the RFO domain propagates to the nuclease domain, leading to proximity-induced dimerization of DUF nuclease domains for activation.

DISCUSSION

Here, we present a paradigm shift for studying the HerA–DUF anti-phage defense system using integrated approaches, emphasizing the intricate interplay between protein structure, assembly, and activity regulation (Figure 7). Leveraging the AlphaFold-predicted structure of DUF as a template, we identified structural homologs of DUF and discovered that DUF shares a similar nuclease domain with Cap4 ¹¹. Our genomic analyses revealed that many DUF nuclease domain-containing proteins, often with varying additional domains, are prevalent in bacterial immune systems, underscoring their importance in anti-phage defense (Figure 1). Building on this insight, we tested DUF’s nuclease activity and found that it is robust only in the presence of HerA. Structural analyses of apo DUF and the HerA-DUF complex revealed that HerA binding drives a dramatic architectural reorganization, transforming DUF from a tightly packed, inactive oligomer into a rearranged state that promotes nuclease domain dimerization and DNA cleavage (Figures 4, 5, and 7).

These findings highlight a cascade of structural events in the HerA-DUF complex. In the apo state, DUF forms a bilayer cylindrical assembly through RFO-RFO interactions with nuclease domains tightly packed as inactive monomers (Figure 4). Upon HerA binding, the HAS domain of HerA engages the RFO domains of DUF, altering the bottom layer’s interactions and dispersing previously tight contacts. This perturbation propagates upward, causing the top layer’s RFO domains to tilt inward. As a result, the nuclease domains dimerize, forming a clamp-like configuration poised for dsDNA cleavage (Figure 5). Remarkably, this rearrangement is achieved primarily through rigid-body rotations, with minimal local conformational changes. Such dramatic architectural remodeling has not been observed in the HerA-Sir2 system or other studied bacterial immune systems, offering mechanistic insights into phage defense strategies. We propose renaming this system the “Proteus” defense system, after the shape-shifting Greek god, to reflect its remarkable structural adaptability.

Nuclease domain dimerization may represent a unifying principle for the activation of this DUF family. The DUF nuclease dimer resembles that of AVAST3 ²², suggesting that proximity-induced dimerization could be a general activation mechanism. This concept mirrors the activation of initiator caspases, such as Caspase-8, where oligomerization of the N-terminal domain increases local concentration and drives proximity-induced dimerization of the catalytic domain ²³.

Our work also challenges the long-held notion that higher-order oligomerization alone is sufficient to activate immune signaling^24,25. Although DUF alone and in complex with HerA both form large oligomers, only the latter is active. This demonstrates that mere oligomerization is insufficient; the precise architecture and conformational rearrangements within these oligomers are essential for activation. These insights may extend to other complex immune systems, where higher-order assemblies must undergo specific remodeling steps to become functionally competent.

Limitations of the study

Our structural and biochemical analyses have provided mechanistic insights into the assembly and function of the HerA-DUF anti-phage defense system. Despite these advances, several questions remain. First, despite the local resolutions of DNA ranging from 3.5 to 5.5 Å, some expected structural features remain invisible, potentially due to the inherent flexibility or heterogeneity of DNA. Second, the nature of phage-derived triggers that activate HerA-DUF is unclear. Identifying these triggers will deepen our understanding of phage-defense initiation. Third, the role of HerA ATPase activity in anti-phage defense remains to be elucidated. Although we established that ATP hydrolysis is important for defense, we have yet to determine how ATP turnover modulates the system’s on/off states. Fourth, while most HerA homologs exhibit helicase activity, the HerA examined here does not. Whether this is a universal feature of the HerA-DUF system or an exception remains unknown. Together, our study provides a mechanistic framework for HerA-DUF activation and opens avenues for investigating the triggers, ATPase roles, and evolutionary diversity of this and related phage-defense systems.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by Tian-Min Fu (Fu.978@osu.edu).

Materials availability

Materials include in this manuscript will be shared upon request.

Data availability

• Accession numbers for DUF in apo state, the DUF-HerA complex, and HerA alone are as follows: (coordinates of atomic models:9C1M, 9C1N, 9C5X, 9C1O, and 9C1X, deposited to Protein Data Bank), and (density map: EMD-45124, EMD-45234, EMD-45126, EMD-45132 deposited to Electron Microscopy Data Bank). Density maps of DUF tetramer, hexamer, and dodecamer, HerA-DUF in complex with DNA were deposited to Electron Microscopy Data Bank with codes of EMD-45129, EMD-34133, EMD-45134, and EMD-45280, respectively.

• This paper does not report original code.

• All data needed to evaluate the conclusions in the paper are present in the paper.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Bacterial strains and culture conditions

The Escherichia coli (E. coli) DH5a strain was used to generate mutations and amplify plasmids. The plasmids were transformed into the E. coli BL21(DE3) strain cultured in Luria broth (LB) medium to recombinantly express proteins including Bacillus ap. DUF4297 and HerA (HMH5848), and HerA-Duf full complex.

Method details

Genomic analysis of DUF4297

The amino acid sequence of DUF4297 from Bacillus ap. HMH5848 was used for protein BLAST (blastp) against standard and experimental databases³⁸. 1,291 sequences were identified with E value 0.05 – 10⁻¹⁴⁶ (0.0), but no CAP4 protein homologs were among the hits. A structural search with Foldseek (3Di/AA) using the atomic coordinates of BsDUF NTD revealed AbpA (UniProt P52127) and Cap4 (UniProt C0VHC9) as the only probable matches with lengths close to the length of the NTD in the AFDB-Proteome or AFDB-SWISSPROT databases ^10,13. Therefore, 48 sequences with probability > 0.7 and similar sequence lengths to the query were randomly selected from the AFDB50 database and were aligned with MUSCLE ²⁶. The resulting alignment was used for PSI-BLAST (nr_pro_Dec14) with eight rounds through the MPI Bioinformatics toolkit as of Jan 2024 (BLOSUM45) ^14,39,40. All collected sequences were filtered to a maximum sequence identity of 95% and 95% sequence coverage with MMseqs2 and clustered on the basis of blastp all-against-all pairwise searches with CLANS until equilibrium at an E value of 1 × 10^{−80 15,27}.

A phylogenetic tree was created from 1009 sequences using iqtree2 (v2.2.2.6-MacOSX) ²⁸, with standard model selection followed by tree inference. 1500 Ultrafast bootstrap analysis with 1500 alerts and nstop 500 were used as cutoffs for tree generation. The tree is configured in a polar tree layout, rooted at the midpoint with the root hidden, nodes set in increasing order, and branches transformed in cladogram style for ease of visualization.

The resulting sequence similarity network was processed with machine learning as previously described by Durairaj, Janani, et. al. 2023 ⁴¹. In short, the similarity network was used as input for GCsnap (v.1.0.10.9) for the analysis of the conservation of the genomic contexts encoding for each of the proteins in the individual clusters ¹⁶. A window of four flanking genes was used. MMSeqs2 was used for protein family clustering ²⁷, and clusters of similar genomic contexts were detected using the operon_cluster_advanced method, which uses PaCMAP (v.0.7.0)²⁹ to project genomic contexts in two dimensions on the basis of their family composition and DBSCAN³⁰ (as implemented in scikit-learn v.1.4.0⁴²) to identify clusters of similar genomic contexts ^43–45. Only families that were found in at least 35% of all genomic contexts were considered to avoid false positive hits. Structural models of DUF4297 homologs were built using AlphaFold (v2.3.2)³² and were visualized in ChimeraX (v1.7.1)³⁷.

Molecular Cloning, Protein expression and purification

HerA and DUF4297 from Bacillus ap. HMH5848 were synthesized into a pET-Duet-1 expression vector with a C-terminal 6x histidine tag, and a pET-28a(+) expression vector with an N-terminal 6x histidine tag, respectively by GenScript, with codon optimization. Mutants of HerA and DUF were created with a Q5 site-directed mutagenesis kit (NEB, catalog # E0554S), and verified with Sanger sequencing. See Supplementary Table 2 for primer and plasmid sequences. HerA and DUF were recombinantly expressed using E. coli BL21(DE3) competent cells (NEB). Bacterial cells were grown at 37 °C in LB supplemented with 100 μg mL⁻¹ ampicillin or 50 μg mL⁻¹ kanamycin until they reached an OD₆₀₀ of 0.8. HerA-DUF complex was co-expressed in LB containing ampicillin and kanamycin. Protein expression was induced by adding 0.3 mM isopropyl β-D-1-thiogalactopyranoside (IPTG), and cultures were further incubated at 18 °C for 16 h. Cells were harvested at 2000 × g for 20 min at 4 °C and resuspended in 20 mM Tris-HCl pH 7.5, 750 mM NaCl lysis buffer. The cell debris was removed after sonication by centrifuging at 40,000 × g for 60 min at 4 °C. The supernatant was loaded onto a column containing Ni-NTA beads (Qiagen, catalog no. 30210) pre-equilibrated with lysis buffer. The column was washed with 10 column volumes (CV) of 20 mM Tris-HCl pH 7.5, 750 mM NaCl, 30 mM imidazole wash buffer, followed by elution using 5 CV of 20 mM Tris-HCl pH 7.5, 750 mM NaCl, 300 mM imidazole elution buffer, in 1 mL aliquots. Eluted protein was pooled and concentrated to 1 mL using an Amicon ^®Ultra 50,000 NMWL centrifugal filter (Millipore, SKU UFC903024), before application to a Superdex 200 10/300 GL Increased or Superose 6 Increase 10/300 GL column (Cytvia, GE28–9909-44, GE29–0915-96) to purify protein to homogeneity. Protein was eluted with gel filtration buffer (20 mM Tris-HCl pH 7.5, 150 mM NaCl) on an ÄKTA go FPLC (Cytvia). HerA and DUF were verified by SDS-Page gel, stained with Coomassie blue. HerA-DUF (D41A) was incubated with blunt dsDNA for 30 minutes before freezing on grids.

Analytical gel filtration analysis

50 μL purified DUF or the HerA-DUF complexes (wild type or mutants) at concentrations of 3 to 5 mg/ml was loaded to a Superose 6 10/300 GL Increased column (Cytvia, GE29–0915-96) for analysis. The gel filtration profiles were overlayed for comparison.

Cryo-EM data collection

Quantifiol R 1.2/1.3 Au 400 mesh 2 nm Carbon (Electron Microscopy Sciences) were glow discharged at 0.2 atm for 30 s and were loaded onto an FEI vitrobot (Thermo Fisher). 3 μL of the sample at 2 mg/ml was applied to the grid before blotting for 4 s and plunge freezing in liquid ethane. Grids were stored in liquid nitrogen prior to screening on a 200 kV Glacios (Thermo Fisher Scientific) equipped with a K3 bioquantum detector (Gatan, Inc.). Grids were again stored in liquid nitrogen until data could be collected using a 300 kV Titan Krios (Thermo Fisher Scientific), also equipped with a K3 detector. For HerA-DUF and apo DUF protein structures, 6879 movies were collected, and for apo HerA, 4,851 movies were collected using data acquisition software EPU (v2.12.1.2782REL), and TIA (v5.0 SP4) (Thermo Fisher Scientific), and 6,216 HerA-DUF (D41A)-dsDNA movies were collected using SerialEM data collection software, all at a magnification of ×81,000, a pixel size of 1.07 Å, at a total dose of 50 e⁻/Ų, and defocus range −0.5 to −2.0 μm.

Cryo-EM data processing

All data processing, including patch motion correction and patch contrast transfer function (CTF) estimation, was performed in cryoSPARC (ver. 4.4.1) ³¹. Initial particles were picked by blob picking from 300 micrographs for generating 2D classes. Representative 2D classes were then selected as templates for template-guided particle picking from all micrographs. After 2D classification, we performed 3D classification and refinement including local refinement. For the DUF dimer, we used C3 and C6 symmetry to do a symmetry expansion followed by local refinement and substantially improved the map resolution. Due to symmetry expansion, the particles number in refinement is higher than the initial particle number. The data processing details are illustrated in supplemental data (Figures S2, S4, and S7).

To determine the local resolution of the HerA-DUF-DNA complex, we first performed local refinement using a mask around the top layer of DUF RFOs and the central pore DNA density from the HerA-DUF(D41A)+DNA density map. C1 symmetry and a low pass filter of 7 Å were applied based off the local resolution estimation from the original DNA density in the central pore ranging from 3.5–5.5 Å resolution. Final resolution range for the top layer of DUF RFOs is 2.4 – 3.0 Å and the final resolution range for Nucleic acid density in the central pore is 2.5 – 5.7 Å. All these tasks were done using cryoSPARC³¹.

Model building and refinement

The initial models of HerA and DUF4297 were predicted with AlphaFold³², and fitted into cryo-EM maps using ChimeraX³⁷. Manual adjustments were made with refinement in Coot (0.9.8.7)³³ and real-space refinement was performed in Phenix (v1.21–5207)³⁴. Top layer DUF NTD density was resolved from noise using DeepEmhancer (v0.14)³⁶ in post-processing. Structural models were validated with MolProbity (v4.5.2)³⁵. All structural images were generated using ChimeraX³⁷.

ATPase assay

The ATPase activity was measured using the ATPase/GTPase Assay Kit (Sigma-Aldrich, #113CB04A30). A working solution of 4 mM solution of ATP was prepared. Varying concentrations of the proteins (50 to 250 nM) were incubated with 1mM of ATP in the buffer of 20 mM Tris pH 8.0, 75 mM KCl, and 2 mM MgCl₂ (included in the kit) in a 40 μl reaction at 37 °C for 30 mins. The reaction was stopped, and the Biotek Synergy HT microplate reader was used to read the reaction products at 620 nm. A phosphate standard curve was generated following the protocol from the kit. The generated standard curve was used to determine ATPase activity.

Nuclease assay

The DNA digestion or cleavage activity of the proteins was performed by incubating 50 nM protein with 10 ng/μl pUC19 plasmid (NEB N3041S), in a 20 μl reaction at 37 °C for varying time conditions in 20 mM Hepes pH 8.0, 75 mM NaCl, 2 mM MgCl₂ buffer, with or without 1mM ATP. The reactions were stopped by adding a 6x purple loading buffer (B7024S NEB). 20 μl sample was run and separated on a 0.5% TAE agarose gel, stained with ethidium bromide. Gels were run at 100 V for 20 min and visualized on Sapphire biomolecular imager (Azure Biosystems).

With the short DNA substrates, 400 nM of proteins were incubated in reaction buffer (20 mM Hepes pH 8.0, 75 mM NaCl, and 2 mM MgCl₂) with 800 nM Cys-labeled nucleic acids substrates (overhang dsDNA, and blunt dsDNA) at 37 °C for 30 mins. The reactions were stopped with EDTA and SDS and samples were separated on 12% DNA PAGE or 2.5% TB agarose gel in TBE or TB buffer at 100 V for 30 min and visualized on Sapphire biomolecular imager (Azure Biosystems).

Helicase assay

The Helicase assay reaction mixtures contained 40 mM Tris, pH 8.0, 1 mM MgCl₂, 15 mM NaCl, 0.1 mM DTT, 1 mg/ml bovine serum albumin, 2 mM ATP, 100 nM DNA substrate, and 2 μM HerA, HerA-DUF, HerA-DUF(D41A) or DDX43 in a final volume of 20 μl. The reaction was incubated at 37 °C for 15 minutes and quenched with EDTA and SDS. The samples were separated on 2.5% TB agarose gel in TB buffer at 100 V for 30 min and visualized on Sapphire biomolecular imager (Azure Biosystems).

Electrophoretic mobility shift assay (EMSA)

For fluorescent protein-DNA binding assay, protein concentrations of HerA-DUF complex, DUF, and HerA between 100 nM to 800 nM were incubated with 800 nM Cy3 -labeled dsDNA substrate in reaction buffer (25 mM Hepes, pH 8.0, 75 mM NaCl) at 37 °C for 15 minutes. The samples were separated on 2.5% agarose gel in 0.5% TB buffer. The gel was imagined on by Sapphire biomolecular imager (Azure Biosystems).

Plaque assay

Plasmids containing HerA, DUF, or HerA-DUF complex or mutants were transformed into BL21(DE3) cells as described in protein expression and purification. A single colony was grown in 2 mL LB containing the appropriate antibiotic, shaking at 220 rpm at 37°C until OD₆₀₀ of 0.1 before inducing with 0.2 mM IPTG and allowing to grow until OD₆₀₀ of 0.4. Then, 500 μl of cultured bacteria was mixed with 14.5 mL 0.5% top agar at 40°C and poured onto plates containing antibiotic and 0.1 mM IPTG. After allowing the top agar to solidify, plates were serially spotted with 2.5 μl T4 phage from 10⁰ – 10⁻⁷. Plates were left to incubate overnight at 30°C.

Mass photometry

To determine the stoichiometry of BsDUF, BsHerA and BsDUF + BsHerA, mass photometry (MP) experiments⁴⁶ were performed using Refeyn TwoMP mass photometer. Glass coverslips (Thorlabs, CG15KH) needed for the measurements were cleaned by rinsing several times alternatingly with water and 100% isopropanol and drying with nitrogen. Culture well-containing silicone gaskets (Sigma, GBL103250–10EA) were cleaned similarly and attached to the glass coverlips. A droplet of index-matched immersion oil (Zeiss Immersol 518 F, UFI: 4Y00-R0DY-1007–3VF3) was added on the mass photometer objective before placing the coverslip on top. All measurements (calibrants and analytes) were carried out in 20 mM HEPES (pH 7.4), 150 mM NaCl, 0.5 mM TCEP. Data was acquired using AcquireMP^™ and analyzed using DiscoverMP^™. Each mass photometer movie was recorded for one minute. For calibration, an equimolar mixture of β-amylase (Sigma: A87811VL) (monomer = 56 kDa, dimer = 112 kDa and tetramer = 224 kDa) and thyroglobulin (Sigma, T9145) (monomer = 330 kDa and dimer = 660 kDa) was used. Calibration was obtained by plotting the ratiometric contrast of each Gaussian distribution against the known masses of calibrants and had an R² > 0.999. Prior to MP measurements, all analytes were diluted to 200 nM. This diluted stock was then used for measurements. The final concentration of samples in gasket wells was 8–12 nM. Masses of the following samples were measured: 1) BsDUF alone, 2) BsHerA alone, 3) BsDUF + BsHerA. Each measurement was carried out in triplicates. Masses were determined using the calibration obtained with β-amylase and thyroglobulin.

QUANTIFICATION AND STATISTICAL ANALYSIS

For the ATPase assay (Figure S4O, Q), the intensities of reaction products were monitored at 620 nm and automatically generated a corresponding data sheet. Each experiment was replicated three times. Excel and GraphPad PRISM were used for statistical analysis. Standard t-tests were used to quantify differences in ATPase activity between HerA and HerA-DUF. Error bars are defined in figure legends.

KEY RESOURCE TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
E. coli T4 Phage	ATCC	11303-B4
Chemicals, peptides, and recombinant proteins
BsDUF	GenScript	N/A
BsHerA	GenScript	N/A
ATP	Sigma-Aldrich	A2383
Ni-NTA Beads	Qiagen	Catalog #30210
6x Purple loading buffer	NEB	B7024S
Critical commercial assays
Q5 Site-directed mutagenesis kit	NEB	Catalog #E0554S
ATPase/GTPase Activity Assay Kit	Sigma-Aldrich	#113CB04A30
Deposited data
HerA-DUF Assembly 1	This paper	EMDB: EMD-45124 PDB: 9C1M
HerA-DUF Assembly 2	This paper	EMDB: EMD-45124 PDB: 9C1N
HerA-DUF Assembly 3	This paper	EMDB: EMD-45234 PDB: 9C5X
Apo HerA	This paper	EMDB: EMD-45126 PDB: 9C1O
DUF (12-mer) - dissociated from HerA	This paper	EMDB: EMD-45132 PDB: 9C1X
Apo DUF (tetramer)	This paper	EMDB: EMD-45129
Apo DUF (hexamer)	This paper	EMDB: EMD-45133
Apo DUF (12-mer)	This paper	EMDB: EMD-45134
HerA-DUF + dsDNA	This paper	EMDB: EMD-45280
Experimental models: Organisms/strains
E. coli DH5α Chemically Competent Cell	NEB	C2987P
E. coli BL21 (DE3) Chemically Competent Cell	NEB	C2527H
Oligonucleotides
Cy3/CCTGATCCAACGGCATACAAGATCGAACCTGGTGAGGATCCCATGGT	This paper	EMSA oligo - 1
TTTTTTTTTCTCACCAGGTTCGATCTTGTATGCCGTTGGATCAGGGATCTTTTTT	This paper	EMSA oligo - 2
GCCGAATTCTACCAGTGAGGATGGACTCCTCACCTGCAGGTTCACCGTTGTATGCCC	This paper	Nuclease overhang substrate – EHJ1
GGTCAACGTGGGCATACAAGGCTGAACCTGCAGGTGAGGAGTCCATGGTCTTCCGTC	This paper	Nuclease overhang substrate – EHJ2
TTTTTTTTTTTTTTTTTAATAACCCGGTTATTTTTTTTTTTTTTTTTTTTTTTTTT	This paper	Nuclease blunt dsDNA substrate - GajF
AAAAAAAAAAAAAAAAAAAAAAAAAATAACCGGGTTATTAAAAAAAAAAAAAAAAA	This paper	Nuclease blunt dsDNA substrate - GajR
Cy3/CCTGATCCAACGGCATACAAGATCGAACCTGGTGAGGATCCCATGGT	This paper	Helicase assay – DE1
TTTTTTGAACTTTTTTTGATCGGTTCGATCTTGTATGCCGTTGGATCAGG	This paper	Helicase assay – DE9
pUC19 plasmid	NEB	N3041S
Oligonucleotides for helicase and cleavage assay, see Table S3
Recombinant DNA
pETDuet-1-HerA-His6x	This paper	N/A
pETDuet-1-HerA-His6x_E25K	This paper	N/A
pETDuet-1-HerA-His6x_H44D	This paper	N/A
pETDuet-1-HerA-His6x_E92K	This paper	N/A
pETDuet-1-HerA-His6x_K149A	This paper	N/A
pETDuet-1-HerA-His6x_K149E	This paper	N/A
pETDuet-1-HerA-His6x_E438/439A	This paper	N/A
pET28A-His6x-DUF	This paper	N/A
pET28A-His6x-DUF_F13/17A	This paper	N/A
pET28A-His6x-DUF_D41A	This paper	N/A
pET28A-His6x-DUF_E59A	This paper	N/A
pET28A-His6x-DUF_K61A	This paper	N/A
pET28A-His6x-DUF_K56/125E	This paper	N/A
pET28A-His6x-DUF_D206K	This paper	N/A
pET28A-His6x-DUF_K246E	This paper	N/A
pET28A-His6x-DUF_K265/314E	This paper	N/A
pET28A-His6x-DUF_K265/314/434E	This paper	N/A
pET28A-His6x-DUF_R337E	This paper	N/A
pET28A-His6x-DUF_D412R	This paper	N/A
Software and algorithms
FoldSeek	van Kempen et al. 10	https://search.foldseek.com/search
BLASTp/PSI-BLAST (nr_pro_Dec14)	Altschul et al. 14	https://toolkit.tuebingen.mpg.de/
MUSCLE	Edgar 26	https://toolkit.tuebingen.mpg.de/
MMSeqs2	Steinegger & Soding 27	https://toolkit.tuebingen.mpg.de/
CLANS	Frickey & Lupas 15	https://toolkit.tuebingen.mpg.de/
Iqtree2 (v2.2.2.6-MacOSX)	Minh et al. 28	https://github.com/iqtree/iqtree2
GCsnap (v1.0.10.9)	Pereira. 16	https://github.com/JoanaMPereira/GCsnap
PaCMAP (v0.7.0)	Wang et al. 29	https://github.com/YingfanWang/PaCMAP
DBSCAN (in scikit-learn v1.4.0)	Ester et al / Pedregosa et al. 30	https://scikit-learn.org/stable/
EPU (v2.12.1.2782REL)	ThermoFischer Scientific	https://www.thermofisher.com/us/en/home/electron-microscopy/life-sciences/learning-center/cryo-em-university/single-particle-analysis-data-collection.html
TIA (v5.0 SP4)	ThermoFischer Scientific	https://www.thermofisher.com/us/en/home/electron-microscopy/life-sciences/learning-center/cryo-em-university/single-particle-analysis-data-collection.html
cryoSPARC (ver. 4.4.1)	Punjani et al. 31	https://cryosparc.com
AlphaFold (v2/v3)	Jumper et al. 32	https://github.com/deepmind/alphafold
Coot (v0.9.8.7)	Emsley and Cowtan 33	https://www2.mrc-lmb.cam.ac.uk/personal/pemsley/coot/
PHENIX (v1.21-5207)	Adams et al. 34	https://phenix-online.org
MolProbity (v4.5.2)	Williams et al. 35	http://molprobity.manchester.ac.uk
DeepEmhancer (v0.14)	Sanchez-Garcia et al. 36	https://github.com/rsanchezgarc/deepEMhancer
ChimeraX (v1.7.1)	Pettersen et al. 37	https://www.rbvi.ucsf.edu/chimerax/
Other
Superose™ 6 Increase 10/300 GL	Cytiva	GE29-0915-96
Superdex 200 Increase 10/300 GL	Cytiva	GE28-9909-44
Quantifoil R 1.2/1.3, 400 mesh, Gold 2 nm Carbon	Electron Microscopy Sciences	Cat#Q4100AR1.3-2nm

Highlights.

• Apo DUF forms an inactive dodecamer with minimal nuclease activity

• HerA-DUF assembles as a 6:12 active complex for efficient DNA cleavage

• DUF oligomerization is mediated by its C-terminal RFO domain.

• Architectural remodeling enables DUF nuclease domain dimerization and activation.